OpenAI schedules removal of nine legacy audio, realtime, and transcription models
OpenAI has notified developers that a set of nine legacy audio, realtime, and transcription model snapshots will be removed from the API on January 20, 2027, giving teams a six-month runway to migrate to current replacements.
What's new
According to OpenAI's API deprecations page, "On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027." The affected list spans nine model families and snapshots, including gpt-realtime, gpt-audio, gpt-4o-audio, gpt-4o-realtime, and their mini variants.
OpenAI has named specific successors for each retiring model rather than leaving migration open-ended: gpt-realtime-2.1 and gpt-audio-1.5 are called out as the recommended replacements for the realtime and audio families respectively. Developers still on the legacy snapshots have until January 20, 2027 to move traffic over before the old endpoints stop serving requests.
Context
This is one of a running series of dated deprecation notices OpenAI has posted through 2026 as it retires older model generations in favor of newer ones — the same deprecations page recorded a separate notice in June covering GPT-5 and o3 shutdowns effective December 11, 2026. Audio, realtime, and transcription models see faster iteration than OpenAI's general text models, since latency and voice-quality improvements land frequently; OpenAI shipped new GPT Transcribe and GPT Live Transcribe models on July 28, just eight days after this deprecation notice went out, reinforcing that older audio-stack models are being displaced by measurably better ones.
Why it matters
For any product built on OpenAI's realtime voice or transcription APIs, this is an operational deadline, not just a housekeeping note — voice agents, live transcription tools, and audio-processing pipelines pinned to the named legacy snapshots need a migration plan before January 20, 2027 or they lose API access outright. Because OpenAI is naming explicit successor models rather than requiring developers to pick a suitable current-generation replacement themselves, the migration path is more concrete than past deprecations, but it still requires testing and configuration changes across whatever integrations were built on the old snapshot IDs.
Corroborating sources
- Developers.openai
https://developers.openai.com/api/docs/deprecations
“On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027.”